In recent, researches have been actively carried out on search engines to search for images that users are looking for. In addition to a search for images that users are looking for, researches have also been carried out on error control as to whether the intentions of the users are fully reflected on the images to be searched. One of methods for searching images with user's intention reflected thereon may be carried out in a way that automatically generates a list of candidate images which may be utilized in making an initial query for an image search. An example of the above-mentioned method is disclosed in Korean Patent Laid-Open Publication No. 2007-0105722, entitled “Automatic Generation Method of the Initial Query Set for Image Search on the Mobile Web,” laid-open on Oct. 31, 2007, which includes: automatically generating a set of initial queries on which a user's preference is reflected using fed-back image relevance information; creating a user profile based on the relevance information; a group of neighbors having a taste similar to the user's preference; and recommending an item that is preferred by the group of neighbors.
However, it is not possible to satisfy fully the need of a user, since the group of neighbors who have a similar taste may have different preferences from the user. In addition, errors between words selected by individual users and associated images may be different from everyone. Therefore, it is needed to calibrate an error between an image intended by a user and a word selected by the user.